Nvinferaudio plugin / mel spectrogram /f_min, f_max

Hardware Platform: Jetson Nano
DeepStream: Version 6.0
JetPack Version: 4.6
TensorRT Version: 8.0.1
Issue Type: questions

when we are using the nvinferaudio plugin, is it possible to put as the audio transform parameter [f_min, f_max] in Hz, so that for the creation of the spectrograms will be used only the specific range of frequencies (from f_min to f_max)? If it’s possible, then how?
like in librosa:
fmin float >= 0 [scalar] - lowest frequency (in Hz)
fmax float >= 0 [scalar] - highest frequency (in Hz)


Sorry for the late response, is this still an issue to support? Thanks

Hi, yes, this issue is still very relevant for us. We have one project, that is working with the wav-files, that contain ultrasound, and it were better if we could create the spectrograms only for the specific frequency range…

Hi @user49080 ,
Sorry for long delay!

This audio transfer is for the audio inference, if we remove some data outside of [f_min, f_max] in Hz, I’m afraid it will cause inaccurate or incorrect inference result.


ok, thanks, but we want to train the model also with the specific frequency range [f_min, f_max] , and than to make inference with the same parameters. So may be it’s somehow possible to set [f_min, f_max] in the audio-transform?